A dead zone fault diagnosis method for an electric drive system based on variable frequency excitation and order tracking

Abstract

The application discloses a kind of based on variable frequency excitation and order tracking electric drive system dead zone fault diagnosis method, belong to electric vehicle fault diagnosis technical field.For the weak dead zone effect characteristics in prior art, easy to be covered by high-frequency switching noise and the problem of frequency blind area in steady-state detection, the application proposes an active diagnosis strategy.The method first identifies motor light load condition, actively reduces inverter PWM carrier frequency to sensitive interval, using increased current ripple to enhance the distortion characteristics caused by dead zone;Subsequently, the motor is controlled to perform acceleration process, and the time-domain non-stationary signal is converted into angular-domain stationary signal using the computed order tracking (COT) technology, eliminating the frequency spectrum ambiguity caused by speed fluctuation;Finally, the local signal-to-noise ratio of the characteristic order is extracted as a quantitative index.The application does not need to increase additional hardware sensors, effectively overcomes the problem of frequency spectrum aliasing and energy dispersion, realizes the high sensitivity, quantitative diagnosis of weak dead zone fault in full speed range.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis technology for electric drive systems of electric vehicles, and specifically relates to a method for highly sensitive quantitative diagnosis of inverter dead-zone effect by actively changing the carrier frequency and combining it with order tracking technology. Background Technology

[0002] In permanent magnet synchronous motor (PMSM) drive systems for electric vehicles, a dead-time must be set to prevent shoot-through between the upper and lower bridge arms of the inverter. However, the dead-time effect can cause output voltage and current waveform distortion, leading to torque ripple and electromagnetic noise, and long-term accumulation can even cause device aging and failure.

[0003] Existing dead zone diagnostic technologies have the following main drawbacks:

[0004] 1. High-frequency noise masking characteristics: In pursuit of quiet operation and control performance, modern electric drive systems typically employ high switching frequencies (e.g., 8kHz-12kHz). Under the masking effect of high-frequency switching noise and fundamental current, the current distortion characteristics caused by microsecond-level dead zones are extremely weak, resulting in a very low signal-to-noise ratio (SNR), making them difficult to extract using conventional methods.

[0005] 2. Blind spots exist in steady-state detection: Existing technologies mostly perform frequency spectrum analysis (FFT) at steady-state speeds. However, due to the modal distribution characteristics of the motor's mechanical structure, the dead zone excitation force at a specific speed may be located at the "anti-resonance point" of the structure, resulting in a weak vibration response and thus missed detection.

[0006] 3. Spectrum blurring caused by speed fluctuations: Under the variable speed conditions of actual vehicle operation, the harmonic frequency caused by dead zone changes in real time with the speed. Traditional FFT analysis will lead to energy leakage and spectrum smearing, making it impossible to accurately locate the fault.

[0007] 4. High hardware costs: Some high-precision methods rely on additional voltage sensors or expensive hardware circuits, which are not feasible for mass production. In automotive-grade applications, adding sensors means increased reliability risks and a significant increase in costs.

[0008] Therefore, there is an urgent need for a dead-zone effect diagnosis method that does not increase hardware costs, can adapt to variable speed operating conditions, and has a high signal-to-noise ratio. Summary of the Invention

[0009] This invention aims to solve the aforementioned technical problems by proposing a dead-zone fault diagnosis method for electric drive systems based on variable frequency excitation and order tracking. This method enhances the physical characteristics of the fault by actively reducing the carrier frequency and eliminates the blind zone by combining it with an order tracking algorithm under acceleration conditions, thus achieving accurate and quantitative diagnosis of dead-zone faults.

[0010] The technical solution of the present invention is as follows:

[0011] A method for diagnosing dead-zone faults in electric drive systems based on variable frequency excitation and order tracking includes the following steps:

[0012] Step S1: Operating Condition Identification and Active Excitation: Real-time monitoring of vehicle operating status. When the motor is detected to be under light load (e.g., torque less than 30% of rated torque), diagnostic mode is entered. The controller instructs the inverter to actively reduce the PWM carrier frequency from the normal operating frequency (e.g., above 8kHz) to a preset sensitive frequency range (e.g., 4kHz-6kHz) to utilize the current ripple to amplify the distortion characteristics caused by the dead zone.

[0013] Step S2: Dynamic Data Acquisition: Control the motor to perform a short acceleration process (Run-up) or utilize the vehicle's natural acceleration process, and synchronously acquire the motor stator current signal (or housing vibration signal) and rotor position / speed signal. In order to capture microsecond-level dead-time distortion, the time step Ts of signal acquisition or simulation must satisfy Ts≤0.1×td, where td is the dead time.

[0014] Step S3: Calculate Order Tracking (COT) resampling: Construct an equal-angle resampling time series using the synchronously acquired rotational speed signal. Convert the time-domain non-stationary signal into an angular-domain stationary signal using an interpolation algorithm. Preferably, the number of sampling points per revolution (PPR) is set between 1024 and 8192 to balance anti-aliasing effect with computational power.

[0015] Step S4: Feature Extraction and Signal-to-Noise Ratio Calculation: Perform a Fourier transform on the angular domain signal to extract the amplitude of the feature order (K=6k×P, where P is the number of pole pairs, usually taken as 6P). Simultaneously calculate the average amplitude of the background noise in the neighborhood of this feature order, and calculate the feature signal-to-noise ratio (SNR).

[0016] Step S5: Fault Diagnosis and Quantitative Assessment: Compare the calculated SNR with a preset threshold. If the SNR exceeds the threshold, a dead zone anomaly is determined to exist; the magnitude of the SNR directly represents the severity of the dead zone effect.

[0017] The beneficial effects of this invention are as follows:

[0018] High sensitivity: By actively reducing the switching frequency (e.g., to 5000Hz), the increased current ripple significantly enhances the dead-zone characteristic signal. Experiments show that compared to the conventional 8000Hz operating condition, the characteristic signal-to-noise ratio can be improved by more than 10dB, solving the problem of weak signals being difficult to detect.

[0019] Full-speed domain without blind spots: It abandons the traditional steady-state detection method and utilizes the "frequency sweeping" effect of the acceleration process, combined with the energy integral characteristics of the COT algorithm.

[0020] It effectively avoids the detection blind zone at specific rotational speeds caused by structural modes, significantly improving the robustness of diagnosis.

[0021] Quantitative grading: The proposed local signal-to-noise ratio index shows a good linear positive correlation with dead time, which can not only qualitatively determine whether there is a fault, but also quantitatively assess the severity of the fault.

[0022] Low cost: Purely software-based, requiring no additional sensors, suitable for deployment in existing production vehicle controllers (VCU / MCU). Attached Figure Description

[0023] Figure 1 This is a flowchart of the method of the present invention.

[0024] Figure 2 This is a comparison chart of characteristic signal-to-noise ratios at different PWM switching frequencies.

[0025] Figure 3 This is a comparison chart of diagnostic results under steady-state and accelerated operating conditions.

[0026] Figure 4 The graph shows the effect of different sampling points per revolution (PPR) on the signal-to-noise ratio.

[0027] Figure 5 This is a linear relationship graph showing the characteristic signal-to-noise ratio as a function of dead time.

[0028] Figure 6 The graph shows the characteristic signal-to-noise ratio variation under different load torques. Detailed Implementation

[0029] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0030] Example 1: Verification of the effectiveness of the frequency conversion excitation strategy

[0031] A high-precision simulation model was established for a permanent magnet synchronous motor system with a pole pair count P=4. Under a fault condition with a dead time td=4μs, the diagnostic performance of different PWM carrier frequencies was compared.

[0032] like Figure 2 As shown, at a conventional switching frequency of 8000Hz, the signal-to-noise ratio (SNR) of the 24th order feature is -3.73dB, indicating that the feature is submerged in noise. When the strategy of this invention is adopted to reduce the switching frequency to 5000Hz, the SNR of the feature is significantly improved to 24.99dB.

[0033] Principle analysis: Lowering the frequency increases the current ripple, which increases the proportion of voltage vector error caused by dead zone distortion in the current waveform, thereby improving the physical strength of the characteristic signal.

[0034] Example 2: Elimination of blind spots under dynamic acceleration conditions

[0035] The diagnostic effects were compared between constant speed (steady state) and acceleration process (transient).

[0036] like Figure 3 As shown, the experiment first tested the diagnostic performance at specific steady-state speeds. Under steady-state conditions of 1000 RPM and 1100 RPM, due to the vibration damping effect of the motor's mechanical structure modes, the dead-zone excitation force failed to excite sufficient structural response, resulting in extremely low signal-to-noise ratios of the detected features (-22.24 dB and -6.11 dB, respectively). The fault features were completely masked by background noise, leading to missed detections.

[0037] It is particularly noteworthy that experiments revealed that while the calculated signal-to-noise ratio (SNR) reached a seemingly high 17.14 dB under steady-state conditions at 1200 RPM, further analysis showed that this data was highly misleading. At this point, the absolute amplitude of the characteristic order was only 0.69, indicating an extremely weak physical signal. The high SNR was merely due to the occasional low background noise (noise floor of only 0.09) under these conditions. This "pseudo-high SNR," relying solely on low noise floor, is highly susceptible to road vibrations or electromagnetic interference during actual vehicle operation and can instantly fail, lacking engineering robustness.

[0038] In contrast, this invention employs an acceleration process of 150-6000 RPM for diagnosis. Utilizing computational order tracking (COT) technology, the algorithm, like an integrator, captures and accumulates the dead-zone energy excited at all resonant points throughout the frequency sweep process. Results show that the extracted characteristic amplitude under accelerated conditions reaches as high as 1.28 (approximately twice the steady-state value at 1200 RPM), and the final measured signal-to-noise ratio is consistently as high as 24.63 dB.

[0039] Conclusion: The dynamic diagnostic strategy of this invention not only effectively overcomes the "frequency blind zone" problem of steady-state detection, but also significantly enhances the physical strength of the characteristic signal through the frequency sweep integration effect, avoiding the "high signal-to-noise ratio artifact" caused by weak signal in steady state, and ensuring the authenticity of the diagnostic results and anti-interference ability.

[0040] Example 3: Optimization of Resampling Parameters

[0041] To determine the optimal signal processing parameters, the impact of different sampling points per revolution (PPR) on feature extraction accuracy was compared.

[0042] like Figure 4As shown, the signal-to-noise ratio (SNR) increases significantly from 1024 to 4096. Specifically, when the PPR is 1024, due to limitations in angular domain resolution, some high-frequency dead-zone distortion information is aliased, resulting in an SNR of 17.40 dB. However, when the PPR is increased to 4096, the SNR reaches the peak value of 26.30 dB in the experimental group. This indicates that at a high resolution of 4096 PPR, the algorithm can most accurately reconstruct microsecond-level current distortion waveforms, completely separating the feature signal from the background noise.

[0043] Subsequently, when the PPR continued to increase to 8192, the signal-to-noise ratio (SNR) decreased instead of increasing (dropping slightly to 26.09 dB). This is because the excessively high sampling rate introduced quantization noise into the original signal and calculation errors from the interpolation algorithm, resulting in an "oversampling" effect.

[0044] Therefore, this invention preferably uses 4096 PPR as the optimal resampling resolution. This parameter selection aims to tap into the ultimate resolution capability of the physical system, ensuring that the faintest early dead-zone fault characteristics can be captured, thereby minimizing the false negative rate and achieving the ultimate diagnostic accuracy.

[0045] Example 4: Load Adaptability and Control Strategy

[0046] like Figure 6 As shown, in the light to medium load range of 1Nm to 10Nm, the characteristic signal-to-noise ratio (SNR) remains high above 20dB, indicating that this range is a "high-sensitivity window" for dead-zone faults. However, when the load increases to 20Nm (heavy load), the SNR drops sharply to the failure zone because the large fundamental current masks the weak dead-zone distortion, and the broadband noise of the system increases under heavy load. In addition, under high current conditions, the magnetic circuit saturation effect causes nonlinear changes in inductor parameters, further obscuring the linear prediction characteristics caused by the dead zone.

[0047] Based on this physical law, this invention designs an intelligent hierarchical control strategy: the vehicle controller (VCU) monitors torque commands in real time, and only activates active frequency reduction and diagnostic programs when the vehicle is in light-load conditions such as coasting, idling, or constant-speed cruising (e.g., <30% of rated torque); while under heavy-load conditions such as rapid acceleration and hill climbing, the automatic locking diagnostic function prioritizes the vehicle's power and safety. This "on-demand diagnostic" strategy ensures both the accuracy of the detection and zero interference with the overall vehicle driving performance.

Claims

1. A method for diagnosing dead-zone faults in an electric drive system based on variable frequency excitation and order tracking, characterized in that, Includes the following steps: Step S1: Operating condition identification and excitation injection: Real-time monitoring of the operating status of the electric drive system. When it is determined that the motor is under a preset light load condition, the inverter is controlled to actively reduce the PWM carrier frequency from the normal operating frequency to the preset sensitive frequency range to enhance the signal distortion characteristics caused by the dead zone effect. Step S2: Dynamic signal acquisition: During the motor's acceleration operation, synchronously acquire time-series signals that reflect the dead-zone effect and rotor speed signals; Step S3: Angular domain resampling: Using the rotational speed signal, the time series signal is processed by computational order tracking (COT) to convert the non-stationary time domain signal into a stationary angular domain signal through equal-angle interpolation; Step S4: Feature extraction and quantization: Perform spectral analysis on the angular domain signal to extract the amplitude of specific feature orders related to the dead zone effect, and calculate the feature signal-to-noise ratio (SNR) by combining the background noise amplitude in the neighborhood of the feature order. Step S5: Diagnosis and evaluation: Compare the characteristic signal-to-noise ratio with a preset fault threshold. If the threshold is exceeded, a dead zone fault is determined, and the severity of the fault is evaluated based on the signal-to-noise ratio value.

2. The method for diagnosing dead-zone faults in an electric drive system based on variable frequency excitation and order tracking according to claim 1, characterized in that, In step S1, the normal operating frequency is 8 kHz or higher, and the sensitive frequency range is 4 kHz to 6 kHz; preferably, the PWM carrier frequency is reduced to 5000 Hz.

3. The method for diagnosing dead-zone faults in an electric drive system based on variable frequency excitation and order tracking according to claim 1, characterized in that, In step S1, the light load condition is defined as the motor output torque being less than 30% of the motor's rated torque; when the motor torque is detected to be greater than 30% of the rated torque, the system automatically locks out the diagnostic function and maintains normal operating frequency.

4. The method for diagnosing dead-zone faults in an electric drive system based on variable frequency excitation and order tracking according to claim 1, characterized in that, In step S2, the time-series signal that can reflect the dead-zone effect includes the motor stator phase current signal or the motor housing vibration acceleration signal; The time step T of the signal acquisition s With inverter dead time t d At least the following relationship must be satisfied: This is to ensure the sampling accuracy of transient distortions caused by dead zones.

5. The method for diagnosing dead-zone faults in an electric drive system based on variable frequency excitation and order tracking according to claim 1, characterized in that, In step S2, the acceleration process covers the speed range of the motor from the low-speed zone to the high-speed zone, preferably covering 10% to 100% of the rated speed range, so as to use the order tracking algorithm to integrate and accumulate the dead zone fault energy in the full speed range.

6. The method for diagnosing dead-zone faults in an electric drive system based on variable frequency excitation and order tracking according to claim 1, characterized in that, In step S3, when performing the order tracking calculation, the number of sampling points per revolution (PPR) is set to a range of 1024 to 8192; preferably, the number of sampling points per revolution is set to 4096 to maximize the spectral resolution and obtain the highest feature signal-to-noise ratio.

7. The method for diagnosing dead-zone faults in an electric drive system based on variable frequency excitation and order tracking according to claim 1, characterized in that, In step S4, the formula for calculating the characteristic order K is: K = 6k × P, where P is the number of pole pairs of the motor and k is a positive integer; preferably, the 6P order when k = 1 is extracted as the principal characteristic order.

8. The method for diagnosing dead-zone faults in an electric drive system based on variable frequency excitation and order tracking according to claim 1, characterized in that, In step S4, the characteristic signal-to-noise ratio (SNR) is calculated as follows: Among them, A target A represents the spectral amplitude at the characteristic order. noise It is the average value of the spectral amplitudes in the left and right neighborhoods of the characteristic order that do not contain the characteristic order.

Images